Croissant: A metadata format for ml-ready datasets

M Akhtar, O Benjelloun, C Conforti… - Advances in …, 2025‏ - proceedings.neurips.cc
Data is a critical resource for machine learning (ML), yet working with data remains a key
friction point. This paper introduces Croissant, a metadata format for datasets that creates a …

Orion: Interference-aware, fine-grained GPU sharing for ML applications

F Strati, X Ma, A Klimovic - … of the Nineteenth European Conference on …, 2024‏ - dl.acm.org
GPUs are critical for maximizing the throughput-per-Watt of deep neural network (DNN)
applications. However, DNN applications often underutilize GPUs, even when using large …

Fastflow: Accelerating deep learning model training with smart offloading of input data pipeline

T Um, B Oh, B Seo, M Kweun, G Kim… - Proceedings of the VLDB …, 2023‏ - dl.acm.org
When training a deep learning (DL) model, input data are pre-processed on CPUs and
transformed into tensors, which are then fed into GPUs for gradient computations of model …

An overview of the data-loader landscape: Comparative performance analysis

I Ofeidis, D Kiedanski… - 2024 IEEE International …, 2024‏ - ieeexplore.ieee.org
The efficiency of Deep Learning (DL) training jobs is critically dependent on dataloaders,
which facilitate the transfer of data from storage to DL-accelerated hardware during training …

Pecan:{Cost-Efficient}{ML} Data Preprocessing with Automatic Transformation Ordering and Hybrid Placement

D Graur, O Mraz, M Li, S Pourghannad… - 2024 USENIX Annual …, 2024‏ - usenix.org
Input data preprocessing is a common bottleneck in machine learning (ML) jobs, that can
significantly increase training time and cost as expensive GPUs or TPUs idle waiting for …

Where is my training bottleneck? hidden trade-offs in deep learning preprocessing pipelines

A Isenko, R Mayer, J Jedele, HA Jacobsen - Proceedings of the 2022 …, 2022‏ - dl.acm.org
Preprocessing pipelines in deep learning aim to provide sufficient data throughput to keep
the training processes busy. Maximizing resource utilization is becoming more challenging …

[HTML][HTML] Data pipeline quality: Influencing factors, root causes of data-related issues, and processing problem areas for developers

H Foidl, V Golendukhina, R Ramler… - Journal of Systems and …, 2024‏ - Elsevier
Data pipelines are an integral part of various modern data-driven systems. However, despite
their importance, they are often unreliable and deliver poor-quality data. A critical step …

tf. data service: A case for disaggregating ML input data processing

A Audibert, Y Chen, D Graur, A Klimovic… - Proceedings of the …, 2023‏ - dl.acm.org
Machine learning (ML) computations commonly execute on expensive specialized
hardware, such as GPUs and TPUs, which provide high FLOPs and performance-per-watt …

PreSto: An In-Storage Data Preprocessing System for Training Recommendation Models

Y Lee, H Kim, M Rhu - 2024 ACM/IEEE 51st Annual …, 2024‏ - ieeexplore.ieee.org
Training recommendation systems (RecSys) faces several challenges as it requires the
“data preprocessing” stage to preprocess an ample amount of raw data and feed them to the …

Intune: Reinforcement learning-based data pipeline optimization for deep recommendation models

K Nagrecha, L Liu, P Delgado… - Proceedings of the 17th …, 2023‏ - dl.acm.org
Deep learning-based recommender models (DLRMs) have become an essential component
of many modern recommender systems. Several companies are now building large compute …